Human Pose and Action Recognition

The paper develops a behavioral analysis pipeline using LUPE to quantify opioid withdrawal behaviors in mice, identifying challenges in capturing rapid behaviors and laying groundwork for standardized quantification.

This paper introduces MATER, a self-supervised framework for reconstructing rodent postures from incomplete data, achieving high-fidelity reconstructions and behavioral classification with minimal supervision.

This paper introduces a 3D tracking technique to analyze social behavior in rodents, revealing new insights into autism and amphetamine effects through high-resolution postural data.

Forestwalk uses machine learning to analyze rodent beam walking, revealing new insights into posture and balance through automated feature extraction and classification.

YOLOv8 detection with SORT event grouping and grid-search tuning quantifies visually distinct behaviours from video across 5 case studies, reaching f1-scores of 0.62 to 0.94.

vmTracking uses virtual markers to improve multi-animal pose tracking by overcoming occlusion and crowding with minimal manual correction.

This paper introduces MarmoPose, a real-time 3D pose tracking system for non-human primates, enabling analysis of social behaviors and closed-loop experiments in marmosets.

This paper improves motion capture accuracy by enhancing sparse keypoints with a more generalizable marker enhancer trained on a larger dataset.

RefineDLC improves DeepLabCut outputs by cleaning, filtering, and interpolating data to produce reliable kinematic analyses across species and recording conditions.

PrecisionTrack provides a platform for automated long-term social behavior analysis in naturalized environments, enabling real-time tracking and social interaction inference in large animal groups.

REVEALS is an open-source GUI for rodent behavior data acquisition using multiple cameras, validated with DeepLabCut for pose estimation in behavioral assays.

This paper introduces SMP, an unsupervised algorithm for decomposing primate behavior into motion motifs, outperforming existing methods and visualizing neural manipulation effects.

This paper evaluates frozen video foundation models for animal behavior analysis, showing they achieve competitive performance across diverse datasets and tasks with minimal adaptation.

SUBTLE maps animal behavior from 3D action skeletons using an unsupervised framework, achieving superior performance in capturing behavioral repertoires and identifying inter-group differences through temporal representation.

FABEL forecasts animal behaviors and locomotion from video using deep learning, showing promise in predicting food interactions and trajectories.

ADPT is a transformer-based pose tracker that reduces tracking drift in animal behavior analysis, achieving high accuracy across species and conditions.

The paper develops a pipeline for realistic 3D macaque body animation using multi-camera markerless motion capture, showing high accuracy and perceptual similarity to real videos.

This paper develops an AI system to quantify social and pathophysiological behaviors in common marmosets using 3D pose estimation and unsupervised analysis.

The paper introduces GlowTrack, a method to generate large-scale training data for markerless motion capture, enabling generalizable deep learning models for animal behavior analysis.

Keypoint-MoSeq fits a generative model that tells tracking jitter apart from real movement, cutting pose time series into unsupervised sub-second behavioural syllables in mice.

The paper introduces SBeA, a framework for tracking and mapping 3D social interactions in free-moving animals, showing it can detect subtle behavioral abnormalities in mice, birds, and dogs.

DeepAction is a MATLAB toolbox that uses deep learning to automate animal behavior classification in video, achieving high accuracy with minimal training data and reducing annotation time.

The paper develops a layered hybrid machine learning workflow to improve the analysis of mouse risk assessment behavior by combining random forest and hidden Markov models.

The paper introduces VAME, a deep learning framework for identifying hierarchical behavioral structures from animal motion data, demonstrating its ability to detect discrete motifs and community-specific usage patterns in mouse models.

BehaviorDEPOT automates behavior classification and analysis in rodents, providing tools for tracking, classification, and experimental setup with reduced costs and improved reliability.

Selfee extracts comprehensive and discriminative features from raw animal behavior videos using self-supervised learning, enabling anomaly detection and temporal analysis.

This paper develops a deep learning tool to automatically score bradykinesia using MDS-UPDRS III protocols, showing significant correlation with gold-standard ratings.

The paper validates DeepLabCut for 3D pose estimation by comparing it to a reference system, showing close agreement in millimeter range.

This paper develops a videography-based method to estimate skeletal pose and joint kinetics in freely-moving rodents, validated by direct limb placement measurements.